Author's Accepted Manuscript

A comparison of optimization methods and

knee joint degrees of freedom on muscle force
predictions during single-leg Hop landings
Hossein Mokhtarzadeh, Luke Perraton, Laur-
ence Fok, Mario A. Muñoz, Ross Clark, Peter
Pivonka, Adam Leigh Bryant

www.elsevier.com/locate/jbiomech

PII: S0021-9290(14)00418-7
DOI: http://dx.doi.org/10.1016/j.jbiomech.2014.07.027
Reference: BM6751

To appear in: Journal of Biomechanics

Accepted date: 27 July 2014

Cite this article as: Hossein Mokhtarzadeh, Luke Perraton, Laurence Fok,
Mario A. Muñoz, Ross Clark, Peter Pivonka, Adam Leigh Bryant, A comparison
of optimization methods and knee joint degrees of freedom on muscle force
predictions during single-leg Hop landings, Journal of Biomechanics, http://dx.doi.
org/10.1016/j.jbiomech.2014.07.027

This is a PDF file of an unedited manuscript that has been accepted for
publication. As a service to our customers we are providing this early version of
the manuscript. The manuscript will undergo copyediting, typesetting, and
review of the resulting galley proof before it is published in its final citable form.
Please note that during the production process errors may be discovered which
could affect the content, and all legal disclaimers that apply to the journal
pertain.
1 A comparison of optimization methods and knee joint degrees of freedom on muscle

2 force predictions during single-leg hop landings

3 Hossein Mokhtarzadeh1, Luke Perraton2, Laurence Fok1, Mario A. Muñoz1, Ross Clark3,

4 Peter Pivonka1, Adam Leigh Bryant2

1
5 Northwest Academic Centre, The University of Melbourne, Australian Institute of Musculoskeletal

6 Science, Melbourne VIC 3021, mhossein@unimelb.edu.au

2
7 Centre for Health, Exercise and Sports Medicine, Physiotherapy, Melbourne School of Health

8 Sciences, Faculty of Medicine, Dentistry and Health Sciences, University of Melbourne , Melbourne

9 VIC 3010, l.perraton@student.unimelb.edu.au

1
10 Northwest Academic Centre, The University of Melbourne, Australian Institute of Musculoskeletal

11 Science, Melbourne VIC 3021, l.fok@student.unimelb.edu.au

1
12 Northwest Academic Centre, The University of Melbourne, Australian Institute of Musculoskeletal

13 Science, Melbourne VIC 3021, mariom@alumni.unimelb.edu.au

3
14 Faculty of Health Sciences, Australian Catholic University, Melbourne VIC, Ross.Clark@acu.edu.au

1
15 Northwest Academic Centre, The University of Melbourne, Australian Institute of Musculoskeletal

16 Science, Melbourne VIC 3021, peter.pivonka@unimelb.edu.au

2
17 Centre for Health, Exercise and Sports Medicine, Physiotherapy, Melbourne School of Health

18 Sciences, Faculty of Medicine, Dentistry and Health Sciences, University of Melbourne , Melbourne

19 VIC 3010, albryant@unimelb.edu.au

20

21

1
22 Word count: 3,258 from introduction to discussion (included)

23

24 Corresponding Author:

25 Hossein Mokhtarzadeh, PhD, GCALL

26 Postdoctoral Research Fellow
27 Australian Institute for Musculoskeletal Science
28 NorthWest Academic Centre
29 The University of Melbourne
30 176 Furlong Road, St Albans Vic 3021, Australia
31 Email: mhossein@unimelb.edu.au
32 Tel.: +61 3 8395 8102
33 Fax.: +61 3 8395 8258
34 Mob.: +61 4 1073 6287
35 Web.: http://aimss.org.au/
36

2
37 Abstract

38 The aim of this paper was to compare the effect of different optimization methods and

39 different knee joint degrees of freedom (DOF) on muscle force predictions during a single

40 legged hop. Nineteen subjects performed single-legged hopping manoeuvres and subject-

41 specific musculoskeletal models were developed to predict muscle forces during the

42 movement. Muscle forces were predicted using static optimization (SO) and computed

43 muscle control (CMC) methods using either 1 or 3 DOF knee joint models. All sagittal and

44 transverse plane joint angles calculated using inverse kinematics or CMC in a 1 DOF or 3

45 DOF knee were well-matched (RMS error < 3o). Biarticular muscles (hamstrings, rectus

46 femoris and gastrocnemius) showed more differences in muscle force profiles when

47 comparing between the different muscle prediction approaches where these muscles showed

48 larger time delays for many of the comparisons. The muscle force magnitudes of vasti,

49 gluteus maximus and gluteus medius were not greatly influenced by the choice of muscle

50 force prediction method with low normalized root mean squared errors (< 48%) observed in

51 most comparisons. We conclude that SO and CMC can be used to predict lower-limb muscle

52 co-contraction during hopping movements. However, care must be taken in interpreting the

53 magnitude of force predicted in the biarticular muscles and the soleus, especially when using

54 a 1 DOF knee. Despite this limitation, given that SO is a more robust and computationally

55 efficient method for predicting muscle forces than CMC, we suggest that SO can be used in

56 conjunction with musculoskeletal models that have a 1 or 3 DOF knee joint to study the

57 relative differences and the role of muscles during hopping activities in future studies.

58 Keywords: Static Optimization, Computed Muscle Control, Musculoskeletal model,

59 hopping, muscle co-contraction, knee joint

60

3
61 Introduction

62 Accurate knowledge of lower-limb muscle forces is important in understanding how muscles

63 function during normal and pathological gait. Reliable estimations of muscle forces can

64 improve predictions of joint contact forces and stresses (Kim et al., 2009) as well as ligament

65 forces (Kernozek and Ragan, 2008; Laughlin et al., 2011; Mokhtarzadeh et al., 2013). A

66 collective understanding of these biomechanical variables can provide insight into the causes

67 or consequences of different joint diseases. For example, accurate knowledge of knee muscle

68 forces can be utilized to improve our understanding of changes in medial and lateral

69 tibiofemoral contact forces after an anterior cruciate ligament injury, which has been

70 suggested to be precursor to knee osteoarthritis (Fregly et al., 2012).

71 Musculoskeletal modelling has recently become a powerful biomechanical tool used to

72 predict muscle forces in which optimization methods are commonly utilized to solve the

73 muscle-moment redundancy problem (i.e. a net joint moment can be produced from an

74 infinite number of muscle force combinations) (Crowninshield, 1978). Static optimization

75 (SO) and computed muscle control (CMC) are two popular optimization methods used for

76 predicting muscle forces and are accessible for use in the freely available musculoskeletal

77 modelling software, OpenSim (Delp et al., 2007; Thelen and Anderson, 2006). SO is an

78 inverse dynamics-based method that partitions the net joint moment amongst individual

79 muscles by minimizing a given performance criterion (e.g. sum of squares of muscle

80 activations) (Erdemir et al., 2007). On the other hand, CMC is a forward dynamics-based

81 approach that utilizes feedback control theory to predict a set of muscle excitations that will

82 produce kinematics that closely match the kinematics calculated from inverse kinematics

83 (Thelen and Anderson, 2006; Thelen et al., 2003). Whilst these methods provide a means for

84 obtaining otherwise unattainable in vivo muscle forces, these predictions are limited in that it

4
 85 is challenging to know how valid or accurate these methods are in predicting individual

86 muscle forces given that no direct measures are available.

87 A previous study has shown that the muscle forces predicted by SO can produce accurate

88 joint contact forces during walking by comparing the predicted contact forces to those

89 measured in a person with an instrumented knee implant (Kim et al., 2009). Previous studies

90 have also shown that SO and CMC produce similar muscle force predictions during walking

91 and running in terms of timing and magnitude (Anderson and Pandy, 2001a; Lin et al., 2011).

92 However, these studies have cautioned against the use of SO for ballistic movements such as

93 jumping as SO may produce muscle activation patterns that are inconsistent with

94 electromyographic (EMG) recordings (Lin et al., 2011). In addition, the ability of SO to

95 predict co-contraction of antagonistic muscles has been criticised because this method

96 excludes muscle activation dynamics. However, several studies have mathematically proven

97 that multi-jointed models containing joints with multiple degrees of freedom (i.e. non-planar

98 joints) can predict co-contraction of antagonistic muscles (Ait-Haddou et al., 2000; Jinha et

99 al., 2006a, 2006b). Given that many past studies have used planar knee joint models i.e. 1

100 degree of freedom (DOF) when predicting muscle forces (Dorn et al., 2012; Fok et al., 2013;

101 Mokhtarzadeh et al., 2013), the current study aims to evaluate the forces generated by the

102 lower-limb muscles using different optimization methods and knee degrees of freedom.

103 Therefore, our study proposes to compare the individual lower-limb muscle force results

104 produced by SO and CMC using both planar and non-planar knee joint models during a

105 ballistic movement (i.e., hopping). We hypothesise that the muscle force results based on the

106 SO method using a 3 degree-of-freedom (DOF) knee joint will be similar to those based on

107 the CMC method from both a 1 and 3 DOF knee joint (H1). On the other hand, we estimate

5
108 that SO results from a 1 DOF knee joint will be significantly lower than the results obtained

109 from other combinations of knee joint types and optimization methods (H2).

110 Methods

111 Nineteen healthy and physically active subjects with no history of knee injury (height = 1.74

112 ± 0.08 m, body mass = 74.2 ± 10.8kg) participated in this study after providing informed

113 consent. Ethical approval was provided by the University of Melbourne’s Behavioural and

114 Social Sciences Human Ethics sub-committee (ethics ID 1136167). Data were collected in

115 the Physiotherapy Movement Laboratory at The University of Melbourne.

116 Participants performed an initial static trial by standing in a neutral position and subsequently

117 completed multiple trials of a single leg hop task. On average two trials per subject were

118 simulated in this study. The distanced hopped was standardised to the participant’s leg length

119 and upper limb movement was standardised by asking participants to fold their arms across

120 their chest. Small reflective markers were mounted on the trunk and both lower limbs of

121 participants. Marker trajectories and ground reaction forces (GRF) were collected

122 simultaneously using a 14 camera Vicon motion analysis system and ground-embedded

123 AMTI force plates. Ground reaction force and marker data were collected at 2400 Hz and 120

124 Hz respectively. Electromyographic (EMG) activity was collected simultaneously with an

125 eight channel Noraxon EMG system (Noraxon USA Inc., Scottsdale, Arizona) sampling at

126 2400 Hz using non-preamplified skin mounted Ag/Cl electrodes (Duotrode, Myotronics).

127 EMG data were collected from the vastus lateralis, vastus medialis, rectus femoris, lateral and

128 medial hamstrings, gluteus medius and medial gastrocnemius muscles of the subject’s

129 dominant leg. A similar filtering method applied in previous studies was utilized for the EMG

130 data (Laughlin et al., 2011; Mokhtarzadeh et al., 2013).

6
131 All musculoskeletal modelling and analyses were performed using OpenSim (Delp et al.,

132 2007), MATLAB and the Edward cluster, a high performance computing (HPC) service, at

133 The University of Melbourne. Kinematic and kinetics data were filtered using butterworth

134 filter with a 4th order, zero-lag, recursive filter with a cut-off frequency of 15 Hz.

135

136 Two different subject-specific musculoskeletal models were generated for each participant (1

137 DOF and 3 DOF knee) by scaling generic models according to body segment dimensions

138 recorded from the static trial. Both models consisted of 92 musculotendon units i.e. Gait2392

139 model in OpenSim. The model with 1 DOF knee had 23 degrees-of freedom while the model

140 with 3 DOF knee had 27 degrees-of-freedom. Musculotendon units were modelled as a three

141 element Hill-type model (Zajac, 1989). The ankle was modelled as 1 DOF joint whereas hip

142 joint consisted of 3 DOF. The maximum isometric force property of each muscle was scaled

143 by a factor of 3 to account for differences in muscle strength between our healthy young

144 adults and the cadavers, which our generic models are based on (Dorn et al., 2012). For each

145 trial and for both models, inverse kinematic analyses were used to calculate joint kinematics

146 by minimizing the distance between model and measured marker trajectories (Lu and

147 O’Connor, 1999) while joint moments were calculated using a traditional inverse dynamics

148 approach. Two optimization methods (SO and CMC) were implemented separately to predict

149 muscle forces to give a total of four approaches for muscle force prediction: (i) SO with 1

150 DOF knee, (ii) SO with 3 DOF knee, (iii) CMC with 1 DOF knee, and (iv) CMC with 3 DOF

151 knee. SO partitions the net joint moments into individual muscle forces by minimising the

152 sum of muscle activations squared at each time instant of the hop-landing cycle (Anderson

153 and Pandy, 2001b). CMC performs a forward simulation to compute a set of muscle

154 excitations that will drive the model to track the experimentally-derived joint angular

7
155 accelerations. Tracking of joint kinematics is achieved through using a proportional-

156 derivative controller while the required set of muscle excitations are calculated using SO

157 (Thelen and Anderson, 2006; Thelen et al., 2003).

158 All analyses were performed over the eccentric landing phase of the task, which encompassed

159 the period from initial foot strike to maximum knee flexion (Mokhtarzadeh et al., 2013). Foot

160 strike was defined as the moment at which vertical GRF just reached above a predefined

161 force (i.e., >10N) and then CMC and SO results were synchronized to account for the time

162 CMC requires to initialize. Landing phase was defined from the time CMC and SO were

163 synchronized to maximum knee flexion angle (0-100%). Using musculoskeletal modelling,

164 nine major lower-limb muscles were compared including vasti (VAS), rectus femoris (RF),

165 gluteus maximus (GMAX), gluteus medius (GMED), hamstrings (HAMS), gastrocnemius

166 (GAS), and soleus (SOL). GMAX and SOL comparisons did not involve EMG. A cross-

167 correlation was performed to compare the similarity in the shape of each muscle force time

168 profile for the four different muscle force prediction approaches. This analysis calculated the

169 time delay required to achieve the maximum unbiased correlation coefficient (R).

170 Specifically, the unbiased correlation coefficient and time delay were calculated by

171 displacing the muscle force profile in time predicted by one method relative to another

172 method (from -100% to 100% of the landing phase) and subsequently, taking the maximum

173 value for the correlation at the time displacement required to achieve this maximum value.

174 For each trial the cross-correlation was performed between the signals resulting from two

175 different methods. The cross-correlation results are a measure of correlation and a measure of

176 time displacement (positive time displacement indicates the first profile has a delay over the

177 second profile, whereas negative means that the first profile has a lead over the second

178 profile). For each comparison, the mean and standard deviation across all trials were

8
179 calculated for the unbiased correlation coefficient and time displacement required (hereafter

180 called the time delay).

181 For each muscle, a normalized root mean squared error (NRMSE) was also calculated

182 between the time-shifted muscle force profiles to compare differences in the magnitude of

183 muscle force predictions. For each muscle, the NRMSE was normalized by the mean force

184 over the entire landing phase and over all muscle force prediction approaches.

185

186 Results

187 All sagittal and transverse plane joint angles calculated using inverse kinematics or CMC in a

188 1 DOF or 3 DOF knee were well-matched (RMS error < 3o) (Figure 1). Residual moments

189 and forces across all participants were also within an acceptable range (RMS < 0.2 BW for

190 residual forces and RMS < 0.05 BW-HT for residual moments) (Figure 2). Finally, muscle

191 force profiles were qualitatively consistent with EMG measurements using all four muscle

192 force prediction approaches (Figure 3).

193

194 Muscle Force Time History Profile

195 The time history profiles were similar for the GMED, VAS and SOL for all muscle force

196 prediction approaches where most comparisons resulted in high correlation coefficients (R >

197 0.7) and time delays of less than 7.5% of the landing phase cycle (Table 1) (Figure 3). The

198 force profiles of GMAX were similar for comparisons which did not involve the combination

199 of CMC with a 3 DOF knee (R > 0.7 and time delays < 8% of the landing phase cycle).

9
200 Biarticular muscles (HAMS, RF and GAS) showed more differences in muscle force profiles

201 when comparing between the different muscle prediction approaches where these muscles

202 showed larger time delays for many of the comparisons (time shift > 8% landing phase) and

203 moderate correlations (0.5 < R< 0.7) for the majority of the comparisons (Table 1).

204 The time profile of SOL was influenced by the choice of knee joint used as small time delays

205 (<3% landing phase) were observed when comparing between the results that used the same

206 knee joint model (Table 1). However, when comparing the muscle force profiles between

207 different knee joint models, large time delays were noticed (>9% landing phase) (Table 1)

208 (Figure 3).

209 Muscle Force Magnitude

210 In general, the muscle force magnitudes of VAS, GMAX and GMED were not greatly

211 influenced by the choice of muscle force prediction method with low NRMSE (< 48%)

212 observed in most comparisons (Table 2) (Figure 3). Bi-articular muscles, HAMS, RF and

213 GAS, generally showed the greatest difference in magnitude between muscle force prediction

214 methods (NRMSE > 112 % for most comparisons). Specifically, the combination of CMC

215 and a 3 DOF knee produced considerably higher HAMS (NRMSE > 163%), GAS (NRMSE

216 > 128%) and RF (NRMSE > 112% BW) forces than in the other optimization-knee joint

217 combinations (Table 3). Similarly large differences in magnitude (NRMSE > 75%) were also

218 observed in SOL where the use of a 3 DOF knee joint resulted in considerably lower SOL

219 force than in a 1 DOF knee (Table 2) (Figure 3). The most similar muscle force magnitude

220 predictions were seen when comparing the predictions from the SO method with a 1 DOF

221 knee and CMC with a 1 DOF knee (NRMSE < 39%).

222

10
223 Discussion

224 The aim of this study was to compare the muscle force predictions given from two

225 different optimization methods (SO and CMC) during a single-leg hopping movement in

226 musculoskeletal models with planar knee joints and models with non-planar knee joints. In

227 general, all four approaches predicted similar muscle force time histories/profiles. However,

228 the magnitude of muscle forces predicted by CMC tended to be higher than SO in most of the

229 major muscles for a given type of knee joint. Also, the use of a 3 DOF knee joint tended to

230 result in larger muscle force predictions than a 1 DOF knee joint when assessing each

231 optimization method independently. However, soleus was an exception to the

232 abovementioned cases as CMC produced less force in soleus than SO for a particular type of

233 knee joint. Furthermore, for a particular optimization method, the use of a 3 DOF knee joint

234 resulted in less force in the soleus than the use of 1DOF knee joint.

235 The results of our study suggest that SO can predict less force output for knee-

236 spanning muscles (HAMS, RF and GAS) when using a 1 DOF knee joint. The reasons for

237 this could be primarily twofold: (1) the SO solution neglects excitation-activation dynamics

238 and, (2) the SO solution only needs to find a combination of muscle forces to satisfy the knee

239 kinematics in one plane (sagittal). However, when more DOFs are included in the knee, the

240 optimization solution must find a combination of muscle forces to match knee kinematics in

241 all three planes. Consequently, greater forces and different muscle force activation patterns

242 may be needed from all knee-spanning muscles to closely match kinematics in all three

243 planes. Nonetheless, it is important to not dismiss SO’s ability to predict co-contraction in a 1

244 DOF knee joint as it still predicted muscle co-contraction, albeit at a lower magnitude. In

245 addition, given that greater co-contraction of knee-spanning is occasionally assumed to be

246 related to greater knee stiffness in clinical practice (Erdemir et al., 2007), one must be careful

11
247 with making conclusions about knee stiffness during ballistic movements since the magnitude

248 of muscle force predictions are influenced by the choice of knee joint and the type of

249 optimization method used.

250 Interestingly, the force in the soleus was substantially lower when using a 3 DOF

251 knee despite it being a uni-articular muscle. It seems that the greater co-contraction of knee

252 spanning muscles predicted when using a 3 DOF knee joint corresponded with a

253 redistribution of the ankle plantarflexor moment from the soleus to the gastrocnemius where

254 there was a substantial decrease in soleus force and a substantial increase in gastrocnemius

255 force. Hence, future studies involving the prediction of ankle plantarflexor muscle forces

256 should carefully consider the choice of knee joint to be used as it will greatly influence the

257 magnitude of forces predicted in these muscles. Furthermore, this finding has implications for

258 the conclusions drawn from previous studies that have used SO to predict ankle muscle

259 forces. For example, one study suggested that SOL has a role in protecting the ACL and

260 based this deduction from the HAMS-to-SOL force ratio and the contribution of the SOL and

261 GAS to the ACL force during single-leg landing (Mokhtarzadeh et al., 2013) whilst another

262 study calculated the contribution of SOL and GAS to the centre of mass acceleration during

263 running at different speeds (Dorn et al., 2012). It is possible that conclusions drawn from

264 these studies could be different if they had used a 3 DOF knee joint (rather than a 1 DOF

265 knee) in their analysis.

266 Interestingly, when SO was used in conjunction with a non-plantar knee joint, the

267 magnitude of muscle force predictions were generally similar to that predicted by CMC in

268 most cases regardless of the type of knee joint. Furthermore, all combinations of optimization

269 methods and types of knee joints produced similar muscle force profiles for the major

270 muscles in terms of their general shape. Given that SO is more computationally efficient

12
271 (approximately five times more efficient) than CMC (Lin et al., 2011), has less preparation

272 time and is more robust than CMC, it seems questionable whether there are justifiable

273 benefits in including activation dynamics as a means of improving muscle force predictions

274 during single-leg hopping. Our study suggests that the use of SO may provide an efficient

275 alternative to CMC whilst yielding similar results - particularly for uni-articular muscles.

276 This study builds upon previous findings showing that similar muscle forces can be predicted

277 for dynamic optimization and SO during walking (Anderson and Pandy, 2001a) and for CMC

278 and SO during walking and running (Lin et al., 2011) by extending the analysis to a more

279 ballistic type of movement (e.g. single-leg hopping). Unlike the current study, these previous

280 studies only used musculoskeletal models with a 1 DOF knee and incorporated tasks that are

281 more cyclic and less physically demanding, which may not be greatly influenced by muscle

282 activation dynamics. Nonetheless, our results were similar to these previous studies in that for

283 a chosen type of knee joint, SO and CMC generally produced similar muscle force time

284 profiles during single-leg hopping (Figure 3). Furthermore, it should be noted that the

285 conclusions from previous studies were founded upon a single trial from one subject whilst

286 our study’s findings were based on multiple trials from multiple subjects, which give us

287 confidence in the conclusions we have deduced.

288 While musculoskeletal models provide a great tool in studying otherwise unattainable muscle

289 forces, this approach does come with limitations. Firstly, it is impossible to know which

290 optimization method and knee joint combination produced the most accurate muscle forces

291 given it is extremely difficult and invasive to measure muscle force in vivo. It is possible that

292 the magnitude and timing of all muscle force predictions are incorrect. In addition, our results

293 for 3DOF models could have been influenced by off-plane (transverse and frontal) kinematic

294 errors (Li et al., 2012). Nonetheless, EMG measurements were qualitatively consistent with

13
295 muscle force time profiles predicted using all four muscle force prediction approaches so that

296 we at least have confidence in the timing of our muscle force predictions (Figure 3).

297 Furthermore, even if the magnitude of muscle forces were inaccurate, we have confidence in

298 the validity of our comparisons given the kinematics were well-matched (Figure 1) and

299 residual forces and moments were small (Figure 2). Secondly, the conclusions obtained from

300 our study apply to single-leg hopping in healthy adults. It is unclear whether the same

301 conclusions can be extended to other ballistic movements such as cutting and jumping.

302 In light of our findings and those of earlier studies, we conclude that both SO and CMC can

303 be used to predict lower-limb muscle co-contraction during hopping movements. However,

304 care must be taken in interpreting the magnitude of force predicted in the biarticular muscles

305 and the soleus, especially when using a 1 DOF knee. Despite this limitation, given that SO is

306 a more robust and computationally efficient method for predicting muscle forces than CMC,

307 we suggest that SO be used in conjunction with musculoskeletal models that have a 1 DOF or

308 3 DOF knee joint to study the relative differences and role of muscles during hopping

309 activities in future studies,; however, there is no agreement on which optimization method

310 can better predict muscle forces during hopping.

311 Conflict of interest statement

312 None of the authors above has any financial or personal relationship with other people or

313 organizations that could inappropriately influence this work, including employment,

314 consultancies, stock ownership, honoraria, paid expert testimony, patent

315 applications/registrations, and grants or other funding.

316

317

14
318

15
319

320 References

321 Ait-Haddou, R., Binding, P., Herzog, W., 2000. Theoretical considerations on cocontraction
322 of sets of agonistic and antagonistic muscles. J. Biomech. 33, 1105–11.

323 Anderson, F.C., Pandy, M.G., 2001a. Dynamic optimization of human walking. J. Biomech.
324 Eng. 123, 381.

325 Anderson, F.C., Pandy, M.G., 2001b. Static and dynamic optimization solutions for gait are
326 practically equivalent. J. Biomech. 34, 153–61.

327 Crowninshield, R., 1978. Use of optimization techniques to predict muscle forces. J.
328 Biomech. Eng. 100, 88–92.

329 Delp, S.L., Anderson, F.C., Arnold, A.S., Loan, P., Habib, A., John, C.T., Guendelman, E.,
330 Thelen, D.G., 2007. OpenSim: open-source software to create and analyze dynamic
331 simulations of movement. IEEE Trans. Biomed. Eng. 54, 1940–1950.

332 Dorn, T.W., Schache, A.G., Pandy, M.G., 2012. Muscular strategy shift in human running:
333 dependence of running speed on hip and ankle muscle performance. J. Exp. Biol. 215,
334 1944–56.

335 Erdemir, A., McLean, S., Herzog, W., van den Bogert, A.J., 2007. Model-based estimation of
336 muscle forces exerted during movements. Clin. Biomech. 22, 131–154.

337 Fok, L. a, Schache, A.G., Crossley, K.M., Lin, Y.-C., Pandy, M.G., 2013. Patellofemoral
338 joint loading during stair ambulation in people with patellofemoral osteoarthritis.
339 Arthritis Rheum. 65, 2059–69.

340 Fregly, B.J., Besier, T.F., Lloyd, D.G., Delp, S.L., Banks, S. a, Pandy, M.G., D’Lima, D.D.,
341 2012. Grand challenge competition to predict in vivo knee loads. J. Orthop. Res. 30,
342 503–13.

343 Jinha, A., Ait-Haddou, R., Binding, P., Herzog, W., 2006a. Antagonistic activity of one-joint
344 muscles in three-dimensions using non-linear optimisation. Math. Biosci. 202, 57–70.

345 Jinha, A., Ait-Haddou, R., Herzog, W., 2006b. Predictions of co-contraction depend critically
346 on degrees-of-freedom in the musculoskeletal model. J. Biomech. 39, 1145–52.

347 Kernozek, T.W., Ragan, R.J., 2008. Estimation of anterior cruciate ligament tension from
348 inverse dynamics data and electromyography in females during drop landing. Clin.
349 Biomech. 23, 1279–1286.

350 Kim, H.J., Fernandez, J.W., Akbarshahi, M., Walter, J.P., Fregly, B.J., Pandy, M.G., 2009.
351 Evaluation of predicted knee-joint muscle forces during gait using an instrumented knee
352 implant. J. Orthop. Res. 27, 1326–31.

16
353 Laughlin, W.A., Weinhandl, J.T., Kernozek, T.W., Cobb, S.C., Keenan, K.G., O’Connor,
354 K.M., 2011. The effects of single-leg landing technique on ACL loading. J. Biomech.
355 44, 1845–1851.

356 Li, K., Zheng, L., Tashman, S., Zhang, X., 2012. The inaccuracy of surface-measured model-
357 derived tibiofemoral kinematics. J. Biomech. 45, 2719–23.

358 Lin, Y.-C., Dorn, T.W., Schache, a. G., Pandy, M.G., 2011. Comparison of different methods
359 for estimating muscle forces in human movement. Proc. Inst. Mech. Eng. Part H J. Eng.
360 Med. 226, 103–112.

361 Lu, T.W., O’Connor, J.J., 1999. Bone position estimation from skin marker co-ordinates
362 using global optimisation with joint constraints. J. Biomech. 32, 129–34.

363 Mokhtarzadeh, H., Yeow, C.H., Hong Goh, J.C., Oetomo, D., Malekipour, F., Lee, P.V.-S.,
364 2013. Contributions of the Soleus and Gastrocnemius muscles to the anterior cruciate
365 ligament loading during single-leg landing. J. Biomech. 46, 1913–1920.

366 Thelen, D.G., Anderson, F.C., 2006. Using computed muscle control to generate forward
367 dynamic simulations of human walking from experimental data. J. Biomech. 39, 1107–
368 1115.

369 Thelen, D.G., Anderson, F.C., Delp, S.L., 2003. Generating dynamic simulations of
370 movement using computed muscle control. J. Biomech. 36, 321–328.

371 Zajac, F.E., 1989. Muscle and tendon: properties, models, scaling, and application to
372 biomechanics and motor control. Crit. Rev. Biomed. Eng. 17, 359.

373

374 Captions:

375 Figure 1: Joint angles during single-leg hopping calculated from computed muscle control

376 (CMC, solid lines) and inverse kinematics (dashed lines) when using a 1 degree-of-freedom

377 knee joint (in black) and 3 degree-of-freedom knee joint (in grey).

378 Figure 2: Residual moments and forces during single-leg hopping calculated from computed

379 muscle control (CMC, solid lines) and static optimization (SO, dashed lines) when using a 1

380 degree-of-freedom knee joint (in black) and 3 degree-of-freedom knee joint (in grey).

381

17
382 Figure 3: Muscle forces during single-leg hopping predicted from computed muscle control

383 (CMC) in a 1 degree of freedom (DOF) knee joint (solid black line) and in a 3 DOF knee

384 joint (solid grey line), static optimization (SO) in a 1 DOF knee joint (dashed black line) and

385 3 DOF knee joint (dashed grey line). Muscle EMG (mean ± std) is shown as shaded regions.

386 BW; body weight

387 Table 1. Cross-Correlation Results (Correlation Coefficient and Time Delay) for different

388 muscles to compare SO, CMC and knee degrees of freedoms.

389

390 Table 2. Magnitude differences (Normalized root mean squared error) for different muscles
391 to compare SO, CMC and knee degrees of freedoms.

392

393

18
394

395

396

397 Figure 1: Joint angles during single-leg hopping calculated from computed muscle control

398 (CMC, solid lines) and inverse kinematics (dashed lines) when using a 1 degree-of-freedom

399 knee joint (in black) and 3 degree-of-freedom knee joint (in grey).

400

401

19
402

403 Figure 2: Residual moments and forces during single-leg hopping calculated from computed

404 muscle control (CMC, solid lines) and static optimization (SO, dashed lines) when using a 1

405 degree-of-freedom knee joint (in black) and 3 degree-of-freedom knee joint (in grey). The top

406 graphs present Pelvic rotations including tilt, list and rotation), and the bottom graphs show

407 pelvic translations i.e. anteroposterior, vertical and mediolateral respectively.

408

409

20
410

411 Figure 3: Muscle forces during single-leg hopping predicted from computed muscle control

412 (CMC) in a 1 degree of freedom (DOF) knee joint (solid black line) and in a 3 DOF knee

413 joint (solid grey line), static optimization (SO) in a 1 DOF knee joint (dashed black line) and

414 3 DOF knee joint (dashed grey line). Muscle EMG (mean ± std) is shown as shaded regions.

415 BW; body weight

416

417

21
418

419 Table 1. Cross-Correlation Results (Correlation Coefficient and Time Delay) for

420 different muscles to compare SO, CMC and knee degrees of freedoms.

Correlation Coefficient (R)

GMAX GMED HAMS RF VAS GAS SOL
SO 1DOF vs. mean 0.77 0.75 0.59 0.68 0.88 0.51 0.72
SO 3DOF std 0.19 0.16 0.18 0.20 0.15 0.15 0.19
SO 1DOF vs. mean 0.85 0.89 0.78 0.80 0.93 0.65 0.92
CMC 1DOF std 0.11 0.10 0.19 0.15 0.06 0.17 0.13
SO 1DOF vs. mean 0.63 0.77 0.55 0.66 0.80 0.64 0.71
CMC 3DOF std 0.18 0.16 0.16 0.19 0.16 0.19 0.19
SO 3DOF vs. mean 0.77 0.76 0.61 0.63 0.81 0.56 0.74
CMC 1DOF std 0.16 0.16 0.19 0.17 0.14 0.18 0.19
SO 3DOF vs. mean 0.70 0.75 0.57 0.67 0.77 0.69 0.65
CMC 3DOF std 0.16 0.17 0.18 0.18 0.19 0.18 0.19
CMC 1DOF vs. mean 0.66 0.81 0.55 0.64 0.80 0.62 0.75
CMC 3DOF std 0.20 0.14 0.14 0.19 0.15 0.18 0.18
Time delays (% landing phase)
GMAX GMED HAMS RF VAS GAS SOL
SO 1DOF vs. mean 5.5 3.0 11.9 2.8 -2.9 -11.2 7.8
SO 3DOF std 21.6 7.7 40.9 21.3 13.2 35.9 23.7
SO 1DOF vs. mean -1.2 0.1 10.4 -4.7 -0.2 -12.9 1.3
CMC 1DOF std 2.6 5.4 28.6 13.2 0.9 22.4 6.7
SO 1DOF vs. mean -4.1 5.6 17.9 -5.2 -2.4 -3.8 5.8
CMC 3DOF std 26.3 23.1 34.0 20.3 6.2 37.7 31.2
SO 3DOF vs. mean -7.0 1.4 -5.6 -1.9 3.5 -12.9 -4.1
CMC 1DOF std 24.0 17.6 41.9 23.6 10.5 33.2 17.0
SO 3DOF vs. mean 0.9 2.1 3.1 -8.3 -0.8 -3.9 -5.0
CMC 3DOF std 15.1 17.3 31.2 23.3 15.1 21.9 19.7
CMC 1DOF vs. mean 0.2 2.1 6.8 -10.5 -1.0 5.1 2.6
CMC 3DOF std 23.5 14.5 35.6 33.2 4.8 38.2 19.1
421 *grey highlights: R > 0.7 or time delay < 7.5% landing phase. Std stands for Standard deviation. The grey
422 highlights represents when the mean correlation coefficient (R) is greater than 0.7 or when the time delay is less
423 than 7.5% of the landing phase. Negative values denote that the first listed method in the comparison best
424 matches the second listed method, when the muscle force time curve is shifted by the reported value. For
425 example, in a SO 1DOF vs. CMC 1DOF comparison for GMAX, a time shift value of -1.2 means that the
426 muscle force time curve predicted using SO 1DOF needs to be shifted 1.2% earlier in the landing phase to
427 produce a correlation coefficient of 0.85.

428

429

22
430 Table 2. Magnitude differences (Normalized root mean squared error) for different
431 muscles to compare SO, CMC and knee degrees of freedoms.

GMAX GMED HAMS RF VAS GAS SOL

SO 1DOF vs. mean 48% 14% 74% 91% 16% 134% 98%
SO 3DOF std 28% 5% 31% 48% 15% 39% 46%
SO 1DOF vs. mean 18% 10% 35% 38% 10% 15% 21%
CMC 1DOF std 5% 3% 23% 21% 5% 9% 18%
SO 1DOF vs. mean 77% 15% 234% 186% 26% 180% 92%
CMC 3DOF std 32% 7% 72% 49% 14% 49% 22%
SO 3DOF vs. mean 38% 18% 55% 72% 19% 129% 87%
CMC 1DOF std 26% 4% 32% 46% 13% 41% 42%
SO 3DOF vs. mean 41% 18% 163% 112% 19% 56% 38%
CMC 3DOF std 20% 8% 78% 41% 9% 52% 30%
CMC 1DOF vs. mean 65% 12% 193% 140% 25% 170% 76%
CMC 3DOF std 33% 6% 64% 53% 12% 43% 20%
432 *grey highlights: normalized root mean squared error < 75%. Std stands for Standard deviation.

433

434

23
Minerva Access is the Institutional Repository of The University of Melbourne

Author/s:
Mokhtarzadeh, H; PERRATON, L; Fok, L; MUNOZ ACOSTA, M; Clark, R; Pivonka, P;
Bryant, AL

Title:
A comparison of optimisation methods and knee joint degrees of freedom on muscle force
predictions during single-leg hop landings

Date:
2014

Citation:
Mokhtarzadeh, H., PERRATON, L., Fok, L., MUNOZ ACOSTA, M., Clark, R., Pivonka, P. &
Bryant, A. L. (2014). A comparison of optimisation methods and knee joint degrees of
freedom on muscle force predictions during single-leg hop landings. Journal of
Biomechanics, 47 (12), pp.2863-2868. https://doi.org/10.1016/j.jbiomech.2014.07.027.

Persistent Link:
http://hdl.handle.net/11343/43810